MarkerDaSharker/amharic-live-translator

# Amharic Real-Time Translator A real-time speech-to-text and translation system for Amharic language, designed to transcribe and translate conversations from Google Meet or microphone input into English. ## 🎯 Features - **Real-time Amharic transcription** using state-of-the-art Wav2Vec2 models - **Automatic English translation** via Google Translate - **Multiple input sources:** - Desktop audio capture (Stereo Mix) for Google Meet calls - Microphone input for direct speech - Phone microphone via web interface - **Dual interfaces:** - Desktop overlay with Pygame for local use - Web-based interface for remote/mobile access - **Advanced audio processing:** - Silence detection and speech thresholding - Adjustable input gain (1x-5x) - Dynamic buffer management - **Three AI model options** with automatic fallback ## 🚀 Quick Start # Python 3.12 recommended conda create -n amharic_translator python=3.12 conda activate amharic_translator # Install dependencies ``` pip install torch torchvision torchaudio --index-url download.pytorch.org pip install transformers deep-translator scipy pygame flask flask-socketio pyaudio ``` ### Desktop Mode (Capture System Audio) ```bash # Capture desktop audio (e.g., Google Meet) python main.py --desktop ``` This will: 1. Load the Wav2Vec2 Amharic model (~1.2GB on first run) 2. Capture audio from Stereo Mix/Desktop Audio 3. Display transcriptions in a Pygame overlay window ### Microphone Mode (Direct Speech) ```bash # Capture from microphone python main.py --mic ``` ### Web/Phone Interface ```bash # Start the web server python server.py ``` Then: - **On PC:** Open `localhost` - **On Phone:** Open `your_pc_ip` **Phone Setup (Android Chrome):** 1. Go to `chrome://flags` 2. Search "Insecure origins treated as secure" 3. Add `your_pc_ip` 4. Enable and relaunch Chrome ## 📱 Web Interface Features - **Microphone selector** - Choose between phone mics (bottom/top/external) - **Vo …